1. Technical Field
The invention is related to devices for digitally encoding analog signals for transmission in a digital communication channel. In particular, the invention is an improvement in variable slope delta modulation coding.
2. Background Art
Continuously variable slope delta modulation (CVSD) coding of signals provides relatively low compression ratios but has the advantage of being very robust to errors in transmission. For this reason, it is an ideal way for coding signals in low power radio-based networks of portable electronic devices. Such networks are sometimes referred to as piconets. In fact, a recently proposed industry standard for such piconets relies upon CVSD coding. Piconets are but one example of the application of CVSD coding. In a piconet according to the proposed industry standard, as many as seven electronic devices may be networked together via radio, specifically using transceivers operating in the license-free 2.45 GHz band. Such portable electronic devices may include a portable (notebook) computer, a cellular telephone, an access port to a local area network, a head set, computer peripherals, (printers, etc.). Of course, the cell phone can provide the piconet access to the internet.
CVSD coding is described, for example, in U.S. Pat. No. 4,783,644 and in U.S. Pat. No. 4,446,565, both of which are incorporated herein by reference. The first aforementioned patent describes CVSD coding of speech signals. A CVSD encoder operates by comparing the input analog signal with a signal reconstructed from the digital output of the encoder. When the input analog signal amplitude is less or greater than the reconstructed signal, the digital output of the encoder for the next clock period is set to one or the other binary value, respectively. The reconstructed signal is produced by supplying the encoded digital signal to an integrator with a continuously variable slope. The continuously variable slope is adjusted to track more closely the input analog signal. Conventionally, the step size of the integrator is fixed. In the proposed industry standard for piconets, the CVSD encoder has been improved somewhat by adjusting the step size depending upon whether the digital output signal binary value changes over a certain number of samples.
Despite the improvement of the step size adjustment in the proposed industry standard version of CVSD coding, it is recognized by the present inventors that the step size does not change sufficiently fast. As a result, in regions where the input signal is of low dynamic range, the encoded digital output represents a poor approximation of the input analog signal. This is because the step size exceeds the input analog signal amplitude, and the digital output signal cannot decrease sufficiently fast to track the analog input signal. The resulting oscillations in the digital output signal can be reduced only by low pass filtering.
Another problem in CVSD signal coding is that the error between the analog input signal and the encoded digital output signal is signal-dependent. The mean and variance of the error are signal-dependent. This effect is referred to as noise modulation which, at lower bit rates, becomes audible and in any case presents a source of signal distortion that reduces system performance. Thus, CVSD signal coding appears to be hampered by inherent limitations on performance that distort the encoded digital output signal. Such distortion manifests itself as higher error rates in the communication channel. Such errors either overwhelm the error correction capability of the communication system, leading to failure, or require more data overhead for error correction which reduces the maximum data rate of the system. However, it has not seemed possible to overcome such problems in CVSD signal coding.
The invention is embodied in a method and apparatus for continuously variable slope delta modulation coding of signals, the apparatus including a thresholder having an analog input and a digital output representing the relationship between a signal amplitude at the analog input and a predetermined threshold. An integrator has an output and one input connected to the output of the thresholder and a second input that receives a step size value, the output of the integrator corresponding to a product of the thresholder output and the step size value. The apparatus further includes an adder having one input that receives an analog input signal that is to be encoded and a second input connected to the output of the integrator, the output of the adder being coupled to the analog input of the thresholder. A step size controller is responsive to an analog signal level related to the analog input signal for varying the step size value in response to variations in the analog signal level. In addition, the apparatus may further include a source producing noise. A noise amplitude controller is responsive to an analog signal level related to the analog input signal for varying the amplitude of the noise in response to variations in the analog signal level to produce noise having a controlled amplitude sufficient to reduce the correlation with the analog input signal of an error between the output of the integrator and the analog input signal. An adder adds the noise having a controlled amplitude to the analog input signal.
With the step size control being responsive to changes in the input signal dynamic range as described above, the problem of distortion at low dynamic range is solved. In addition, however, the present invention also solves the problem of noise modulation or the dependence of the noise or error on the input signal. This latter problem is solved by adding pseudo-random noise to the analog input signal. The amplitude of the pseudo-random noise is controlled relative to the amplitude of the input signal so that it is relatively small. Specifically, in one implementation it is equal to the least-significant bit of the desired audio resolution. This level of noise is sufficient to transform the signal-dependent error into signal independent error. The effects of signal-independent error are much more benign to the ear (than signal-dependent noise) because it is random noise. At high bit rates, this noise may be below the audible threshold. It is a discovery of the invention that it is advantageous in CVSD coders to reduce (or eliminate) the correlation of the error with the input analog signal at the expense of increasing uncorrelated (random) error.